Kano Survey

By Gojko Adzic · Published 20 September 2024 · Updated 29 September 2026

The Kano Survey is a research tool for discovering customer expectations related to product features, according to the Kano Model. Kano surveys can be useful to help understand customer needs and preferences, choose between alternative options, uncover hidden customer expectations about basic product functions and potential exciters, and help with prioritising product development work. They can also help to prevent over-engineering by identifying unnecessary features, or improvements that would not contribute to customer satisfaction. At the same time, the Kano Survey is imprecise, and should not be used as the key factor for prioritization, but as a supplemental tool to evaluate assumptions and direct further research.

This article explains how to create Kano surveys, how to score and interpret the results with Kano analysis, and several methods for visualizing and statistically evaluating the confidence of the results. It also provides some guidelines on applicability and limitations, and when to use Kano surveys.

• Kano Model
  • Kano model categories
  • Alternative names for Kano model categories
  • An example of Kano Model categorization
• Kano questionnaire
  • How to formulate questions for Kano surveys
  • Kano answer option variants
  • Structural questionnaire variants
• Kano method origins and influences
• Interpreting Kano survey results
• Quantifying Kano surveys
  • Pick the most common category
  • Report a combination when unclear
  • Report satisfaction coefficients
  • Use continuous scoring
  • Use the configuration index from A-Kano
  • Report the full distribution
  • How Votito interprets Kano survey results
• Visualising Kano results
  • How Votito shows Kano results
• How many people are required for a Kano survey?
• Applicability
• Kano Model Limitations
  • Classification changes over time
  • One category rarely fits all customers
  • A single person might fit into multiple categories
  • The model does not consider the magnitude or limits of performance
  • The survey is unreliable
  • The model might expect durable properties

Kano Model

The Kano Model categorizes product functions or features based on how customers perceive them. The basic assumption of the Kano model is that implementing different features impacts user satisfaction in different ways. For some features this relationship is linear (more is proportionally better), and for some it’s asymmetric and non-linear. Understanding this relationship through Kano modeling can help to guide how much to invest in which feature type.

The Kano model is named after Noriaki Kano, who introduced with colleagues in 1984 with the goal of challenging the then established notion that any improvements to key product areas are likely to increase customer satisfaction. In 1997 Kano received the Deming Prize for Individuals.

Kano model categories

In the Kano model, there are three main categories of product qualities or functions, and three additional categories.

The main categories are:

1. Must-be: Aspects that customers take for granted. When these aspects are done well, the customers are not especially satisfied, as that is simply what they expect. When done badly or missing, the customers are dissatisfied. Increasing the implementation level of product features in this class will not, in itself, increase user satisfaction.
2. One-dimensional: reflecting competitive product aspects that customers value. Customers appreciate them when fulfilled. When these functions are not present or fail to meet the expectations, the customers are dissatisfied. Features in this category have a linear relationship with user satisfaction (a straight line on the Kano diagram). The more the feature is implemented, the more satisfied the users become.
3. Attractive: Aspects that customers value when fulfilled, but do not cause dissatisfaction when not fulfilled. Such features are not expected, but their presence can greatly increase user satisfaction.

The additional classes are:

1. Indifferent: the presence or absence of a feature does not affect user satisfaction
2. Reverse: features whose presence decreases user satisfaction, or whose absence increases it
3. Questionable: features where the users indicated conflicting answers, such as liking both the presence and the absence of a feature, or disliking both. Kano’s original paper calls them sceptical responses, as this category says more about the interview participant or response than the actual feature.

Kano analysis uses the model to classify different types of features and help decide on how much to invest in them, including basic functions and qualities which customers take for granted, performance functions where customers appreciate improvements, and delighter functions which customers do not expect but value. In that sense, the Kano model is “used to confirm and categorize the customer requirements discovered through voice-of-the-customer visits”.

Alternative names for Kano model categories

Kano’s original paper was in Japanese, but provided English names for its “quality elements”: attractive, one-dimensional, must-be, indifferent and reverse. Later authors introduced many alternative terms, so there are several competing naming systems for the Kano categories. Here are some of the most popular alternatives:

• Must-be: basic, expected, minimum, flat, hygiene factors, threshold, dissatisfier
• One-dimensional: performance, critical, key, satisfier, unitary
• Attractive: exciter, delighter, motivator, satisfier, value-added
• Indifferent: low-impact, secondary, unimportant

Kano himself, writing in English in 1995, called attractive quality “attractive, or exciting” quality, must-be quality “expected” quality, and one-dimensional quality “unitary” quality.

Note that “satisfier” and “dissatisfier” tend to be used inconsistently across different literature. The original terms come from Herzberg’s motivator-hygiene theory, where a motivator is “sometimes called a ‘satisfier’” and a hygiene factor is “sometimes referred to as a ‘dissatisfier’”. Applied to Kano categories, the words have been used in two conflicting ways:

The 1993 Special issue on Kano’s methods notes that “some companies” called must-be elements “dissatisfiers” and some called one-dimensional requirements “satisfiers”.

The word “delight” does not appear in Kano’s original paper. Shoji Shiba, Alan Graham and David Walden’s 1993 textbook reports that some companies “call these attractive elements ‘delighters’”, because “they do not dissatisfy if absent but they can delight when present”. The 1993 special issue repeats this. Shiba’s book uses “customer satisfied” as the label at the top of the Kano diagram’s vertical axis. In the Special issue, Fred Pouliot relabels it “Satisfaction (Delight)”, assuming that “like” means a very strong like, “which in the limit becomes complete customer ‘delight’”.

Although the Kano model is often associated with delighters in modern web usage, in this case calling attractive features “delighters” is just a label, not a finding. We did not find any studies that connect Kano categorization to measuring delight as an emotion.

Using relatively generic customer expectations from mid-2020s, with abundant cheap cloud storage and professional consumers used to accessing content from multiple devices such as phones and laptops, a Kano model example for a note-taking application would categorise features into groups like this:

• Must-Be: Data persistence and integrity (customers would expect any changes to their notes to automatically be saved to their local device, and be available even in case of a device crashing due to battery running out or an accidental exit from the application). Unicode support for capturing notes in a local language (customers would not expect problems with missing font glyphs).
• One-Dimensional: Quick synchronization to other devices (For customers using the application on multiple devices concurrently, synchronization delays of a few seconds are probably acceptable. Delays measured in minutes would cause dissatisfaction. Almost real-time synchronization would be appreciated. In general, faster is better.) Search speed and accuracy, especially with lots of notes or very long notes (Searching should not take more than a few seconds, and only the matching entries should be shown so that a user can quickly overview them. Showing non-matching entries, missing matching entries or very slow search would cause dissatisfaction.)
• Attractive: Recovery from unwanted destructive actions by the user (for example, being able to go back in history and retrieve a part of a note that a user deleted, but later realised they still need).
• Indifferent: In the context of professional users, being able to set the background image to look like rustic paper or similar cosmetic changes.
• Reverse: Unbundled pricing (e.g. asking the users to pay separately for different categories of storage, synchronization and functions). While some users might appreciate being able to get a cheaper price by only paying for the functions they need, others might be annoyed by constantly hitting limits and being asked to upgrade. A simple pricing scheme, potentially with a small number of tiers to allow different storage capacity, would be expected by most consumers used to similar applications on the market.

The most common method to classify items using the Kano model is a structured questionnaire, typically called the Kano Survey or Kano Method. The questionnaire evolved through usage and there are several popular variants of how to phrase the questions and answer options. The Kano questionnaire asks participants to select their attitudes towards a product feature from two opposite perspectives. The two Kano questions are commonly called:

• Functional: how they would feel if the feature was present (implemented)?
• Dysfunctional: how they would feel if the feature was absent (not implemented)?

Here is an example of the questionnaire with the functional and dysfunctional forms:

The two answers are combined into a category using an evaluation table. The combination of responses helps to spot contradictory answers, such as liking both the presence and the absence of a feature.

Kano designed the paired questions and the evaluation table as a way to test his theory of quality, without giving it a specific name. The name “Kano Questionnaire” and its standard English translation were published in Shoji Shiba, Alan Graham and David Walden’s book A New American TQM.

Note that the original questionnaire presented the dysfunctional question before the functional one, and included an “Other” option for uncertain users. In modern usage, it’s more common to put the functional question first, and exclude the “Other” option.

In a survey for NASA, Lee and Newcomb shuffled the question pairs so that the functional and dysfunctional forms of the same question were not next to each other. We could not find any methodical evaluation of whether reordering makes a difference.

How to formulate questions for Kano surveys

When formulating Kano questions, Kurt Matzler and colleagues suggest that it is critically important to frame them from a customer perspective, not from a technical perspective:

The “voice of the customer” is a description of the problem to be solved from the customer’s viewpoint. If one asks about the technical solutions of a product, it can easily happen that the question is not correctly understood. The customer is not interested in how but which of his problems will be solved. In addition, if the solution to the technical problem is already provided in the formulation of the question, the engineers’ creativity might well be restricted in the field of product development at a later date.

– Kurt Matzler, Hans H. Hinterhuber, Franz Bailom and Elmar Sauerwein, How to delight your customers

The authors of Kano’s Methods for Understanding Customer-defined Quality, adapting content from the Concept Engineering manual by Gary Burchill, Diane Shen and others, suggest the following guidelines:

• You may need to step down the ladder of abstraction to construct a clear question, but avoid straying from the original intent of the customer requirement statement
• Beware of polar wording in the question pairs, and prefer multivalued orientation
• Don’t try to cram several thoughts into one question. Instead, use multiple questions but bear in mind the need to keep the survey to a manageable length
• Make sure the questions are in customer terms, not development terms (benefits, not features)

Burchill makes a related point on the “polar wording” guideline, saying that “To write a 0/1 pair of questions is easy, but when you develop question pairs that are anchored on opposite ends of the scale, the responses you receive are often questionable”.

Josip Mikulić and Darko Prebežac recommend asking about the “benefits to be expected through the provision of an attribute rather than the provision” itself.

Kano answer option variants

There is no specific standard way to formulate Kano questions or response options, and different researchers used different forms. The original Japanese paper offered six answers, which loosely translate as “I like it”, “It is a matter of course”, “I feel nothing about it”, “It can’t be helped”, “I don’t like it” and “Other”.

Shoji Shiba’s 1993 English translation phrases the options as longer statements, such as:

1. I like it that way
2. It must be that way
3. I am neutral
4. I can live with it that way
5. I dislike it that way

Several other alternatives were published in Kano’s Methods for Understanding Customer-defined Quality. One alternative, which the paper claims “seems to differentiate better among the responses” (without any proof), is:

1. I enjoy it that way
2. It is a basic necessity or I expect it that way
3. I am neutral
4. I dislike it, but I can live with it that way
5. I dislike it, and I can’t accept it

Reinhart Richter reports that BBN’s international customers were confused by the ranking of the first two options from Shiba’s translation, and people ‘thought that “It must be that way” was much stronger than “I like it that way”’. Richter’s team at BBN thought that phrases such as “I like it that way” were too ambiguous, and used the following wording instead:

1. This would be very helpful to me
2. This is a basic requirement for me
3. This would not affect me
4. This would be a minor inconvenience
5. This would be a major problem for me

The same source also includes Fred Pouliot’s structure, where the options list the attitudes directly:

1. I like it
2. I expect it
3. I’m neutral
4. I can tolerate it
5. I dislike it

Chris Chapman and Mario Callegaro used the Pouliot version, but with “I feel neutral” instead of “I’m neutral”.

Kano himself offered an alternative wording for the questionnaire in 1995, with the question ending in “how do you feel?”:

1. Favorable
2. Expected
3. Neutral
4. Tolerant
5. Unfavorable
6. Others

The different phrasing options seem to reflect researcher preferences and the cultural or linguistic understanding of the respondents. We have not been able to find any study that has compared these phrasings in a statistically valid way.

Structural questionnaire variants

One of the issues with the standard questionnaire is that it does not reflect the magnitude of a category. A mildly attractive and a very attractive feature end up in the same bucket. To avoid that, and to help resolve ties, you can add an importance question for every requirement. For example, ask the participants to rate each feature on a scale from “Not at all important” to “Extremely important”.

Brand and Rese propose a “Dual Response” Kano method, with follow-up questions triggered by choices that would rank a response as indifferent.

Kano himself proposed a simpler three-level variant in 2001, asking respondents to select only between satisfied, neutral and dissatisfied, using a simple 3 × 3 evaluation table to interpret the results.

Witell and Löfgren compared the five-level questionnaire with the three-level one and with two other alternative ways to score features. The results for the alternatives, at best, agreed with the five-level questionnaire only for two out of the nine tested attributes. The three-level version pushed more attributes into the one-dimensional category than the five-option version. Witell and Löfgren still recommend the five-level questionnaire, although the study could not show which classification is right.

Kano’s model was originally published (without that specific name) in Attractive Quality and Must-Be Quality, by Noriaki Kano, Nobuhiko Seraku, Fumio Takahashi and Shin-ichi Tsuji, in the April 1984 issue of the journal of the Japanese Society for Quality Control. The paper is in Japanese, with an abstract in English.

According to the authors, the model was inspired by the two-factor theory of workplace motivation, introduced by Frederick Herzberg and colleagues in 1959. Kano and Takahashi first presented it in 1979 as the “M-H property of quality” (motivator-hygiene). In 1981 Kano renamed it “attractive quality” and “must-be quality”, because the initial name was not easy to relate to. The paper also mentions Japanese precursors, such as Shigeru Mizuno’s plus and minus quality.

The paper tested the theory with two questionnaire surveys, including one where 540 consumers rated five quality elements of clocks. The clock survey was used by the Rhythm Watch Co., Ltd. (リズム時計工業) to plan a new product, the “Quartz Clock Mecha-Skeleton”. A chart in the paper shows the new clock’s cumulative sales index, over its first year and a half, at roughly three times the average of existing similar products. The text itself only calls this a good start, and credits company-wide quality control as well as the survey.

Swan and Combs split product performance into instrumental (the physical product) and expressive (psychological) dimensions in 1976, a few years before Kano’s work was published. Their work was also based on Herzberg’s theory of motivation. Swan and Combs took the critical-incident interview technique from Herzberg, and they found that dissatisfying items mostly involved failures of the physical product, while satisfying items involved a mix of physical and psychological, discovering an asymmetry similar to the model Kano later published. Kano’s original paper does not reference Swan and Combs, and only names Herzberg as the inspiration.

Kano’s method was translated to English by Shoji Shiba, who included it in the book A New American TQM written together with Alan Graham and David Walden. The translation also appeared in the Center for Quality of Management special issue on Kano’s Methods for Understanding Customer-defined Quality, which adapts the book’s text and expands it with other reports. This version classified customer requirements rather than quality elements, described features as “functional” or “dysfunctional”, asked about the functional form first, and reduced Kano’s six answer options to five by dropping “Other”. The categories and the 25 cells of the evaluation table stayed the same, except that Kano’s “sceptical” answers became “questionable”. The special issue also added the methods for continuous scoring and Better and Worse coefficients, which could be used for statistical analysis.

Shiba’s translation and work influenced Kano, who in 1995 presented a sample questionnaire “for classifying customer requirements”, recommended Shiba’s book and the special issue, and noted that American quality professionals had introduced his idea “as exciting quality or the Kano model”.

The model has since generated a large research literature. A 2013 review counted 147 research papers published between 1984 and 2012, 124 of them empirical studies, and concluded that the “field in general has come to a standstill”.

Kano analysis usually interprets the survey results by scoring each pair of responses according to an evaluation table. Similar to how there are multiple ways of phrasing questions, there are multiple evaluation tables (without any strict proof of which one is better or worse). Most interpretation tables agree on the key categories.

Shiba’s translation of Kano’s method from Japanese uses the following table:

For example, if customers expect a feature to be present and dislike it when it is not present, the feature is “must-be”, marked as M.

Inconsistent responses (both positive, or both negative) are “questionable”, marked as Q in the rating table. They are usually removed from the data set for analysis. Kano calls these “sceptical” responses, suggesting that the question was perhaps not easy to understand or badly worded. Matzler and colleagues add that the respondent may have ticked a wrong answer by mistake.

Several evaluation tables are in use, differing on what they declare questionable. Three are documented in the research literature:

Kano original paper also had an extra row and column hold “Other” answers, which are counted but not classified. More modern usage generally does not include the “Other” option.

Fred Pouliot proposed marking two more cells as questionable. He argued that answering “expect” to both questions, or “tolerate” to both, is contradictory, and that such responses should be questionable and not indifferent.

Lee and Newcomb, in a survey for NASA’s microgravity research programme, treated five potentially contradictory choices as questionable:

Löfgren and Witell scored one Swedish survey of 708 consumers about 24 packaging attributes with all three tables, and found the classifications were similar “in 96 percent of the cases”.

There is no universally accepted way of quantitatively scoring Kano surveys across multiple participants. There are several methods in use, fitting different situations. Here are the main options, some with a Kano analysis example from published research.

Pick the most common category

Kano suggested picking the most common category for an individual feature in Attractive Quality and Must-Be Quality. The results present each feature with a single category, without confidence intervals or significance tests. This is by far the most popular option. Witell reviewed 147 Kano studies and reported that almost all of them “simply count the number of responses in a certain quality dimension”.

This method is simple, but for contested options it may not reflect the nuances of different choices. Here is a Kano analysis example from Kano’s original paper: one function for the TV set was rated as “must-be” by 413 respondents and as “one-dimensional” by 412, out of 876. The winning category had one vote more than the second, and from a statistical perspective on that population they should be indistinguishable.

A single category can also reflect a minority choice. For one feature in Chapman and Callegaro’s research, the most common category covered only 28–33% of the respondents.

Several authors in Kano’s Methods for Understanding Customer-defined Quality suggest strategies for dealing with situations where a feature fits almost equally into multiple categories:

• The survey population might include multiple customer segments, so split the data and analyse the segments separately.
• Follow up with customers for additional insight.
• Select the classification that would have the greatest impact on the product, in the order must-be, one-dimensional, attractive, indifferent (M > O > A > I).
• If questionable is one of the high score candidates, clarify the question and re-run the analysis.

Report a combination when unclear

In a survey for NASA’s microgravity research programme, Lee and Newcomb reported two extra numbers for each requirement:

• Category strength: the difference in percentage points between the most common and the second most common category.
• Total strength: the combined share of attractive, one-dimensional and must-be answers, which Lee and Newcomb used as “the measure of the interest in that issue”. NASA acted on requirements with a total strength above 60%.

Lee and Newcomb required a difference of at least 6 points between the top two categories for a clear choice, or they reported the requirement as a combination of categories. In one Kano analysis example from their survey, an item scored 43% one-dimensional and 39% must-be, and because the difference was less than 6 points, it was reported as a combination.

Both numbers are easy to calculate, and they make ties visible instead of hiding them. The problem with this approach is choosing the right threshold.

Instead of a fixed threshold, some researchers test whether the two largest shares differ significantly. Löfgren and Witell used a t-test for this (see Kano’s Theory of Attractive Quality and Packaging and Classification of quality attributes). Brand and Rese use the “Fong test”, pointing to Douglas Fong’s 1996 paper “Using the Self-Stated Importance Questionnaire to Interpret Kano Questionnaire Results”. (We could not find a copy of Fong’s paper, so we could not check the test against its source, this is the one big gap in the article). From the Brand and Rese paper, it seems to be the standard pooled two-proportion z-test at a 90% confidence level. That test is meant for comparing two independent groups of respondents, so it is not applicable to Kano surveys. In a Kano survey, both counts come from the same respondents, and each respondent falls into only one category, so the formula underestimates how much the gap varies and declares a winner too often. To understand if the two main categories are significantly different, the McNemar test would be better, as it compares two properties measured on the same respondents, so if fits well comparing two classifications derived from the same survey.

Report satisfaction coefficients

Mike Timko suggested reporting “Better” and “Worse” coefficients, to help compare features against each other on a two-dimensional scale.

His original formulas define them as:

• Better = (A + O) / (A + O + M + I)
• Worse = -1 * (O + M) / (A + O + M + I)

Matzler and colleagues called them customer satisfaction (CS) coefficients, labelling individual formulas “Extent of satisfaction” and “Extent of dissatisfaction”.

For questionnaires that include importance ranking, Timko suggested multiplying both coefficients by the average importance score, so that “potential requirements judged less important” move towards the indifferent corner.

The precision of the coefficients is unclear. Timko suspected that, because “the resolution of the Kano data is so low”, “the statistical uncertainty of the averages is quite large (tens of percent). Don’t count on these numbers to differentiate between close results.”

Use continuous scoring

William DuMouchel proposed converting individual answers into a continuous numeric score:

• Functional: -2 (Dislike), -1 (Live with), 0 (Neutral), 2 (Must-be), 4 (Like)
• Dysfunctional: -2 (Like), -1 (Must be), 0 (Neutral), 2 (Live with), 4 (Dislike)

The asymmetry is deliberate, and is intended to stop weak reverse and questionable answers from pulling the average too far.

When the importance ratings are included in the survey, DuMouchel suggests multiplying the score by the importance rating in individual responses.

Continuous scores allow the survey to be analysed using statistical methods and reduce the loss of information from grouping together twenty-five different answer groups into six categories. Continuous scoring also helps to handle cases when there is no dominant answer.

Qianli Xu, Roger J. Jiao, Xi Yang, Martin Helander, Halimahtun M. Khalid and Anders Opperud adapted the DuMouchel scoring into a Kano analysis model, which creates a “configuration index” that reflects the probability for a feature to be included in the product, as a combination of the “importance index” and the “satisfaction index”. They called their model the “Analytical Kano”, or A-Kano.

The A-Kano model multiplies DuMouchel scores by the importance rating and then normalises to plot on a graph as a vector. The importance index is the magnitude of the vector (the distance from the origin) and the satisfaction index is the angle from the X-axis. Based on the thresholds of the angle and distance, a feature gets assigned a category: features closer to the origin than 0.5 are indifferent. Beyond that distance, the angle decides the category: must-be up to 25°, attractive above 65°, and one-dimensional in between. The authors warn that “these threshold values may be problem-specific”, so you should not consider them as generally valid, but instead fit your own coefficients based on your data.

The “configuration index” probability is calculated from those two factors using the following formula:

The authors then propose using a mathematical optimisation algorithm to pick the best combination of options, maximising the total index of the chosen features divided by the time it takes to assemble them.

This Kano analysis model is widely cited in academic literature but we were not able to find any proof so far that it predicts customer behaviour or preferences better than the others. The A-Kano method is presented with a case study from car manufacturing, and likely does not apply to digital product development where inputs (such as the duration required to produce something) are notoriously difficult to estimate (see the Cost of Delay article).

Report the full distribution

Chapman and Callegaro advised against using the common Kano scale or continuous scoring, because the answer scale mixes several dimensions. Instead, they recommend reporting the full set of categories, not only the winner, as they claim that individual choices are not reliable and many do not stay the same after a retest.

How Votito interprets Kano survey results

For Votito, we’ve chosen to follow Chapman and Callegaro’s recommendation, and show the share of responses in each category, sorted from the most frequently selected choice to the least one. Each share comes with a confidence interval, at 95% by default (90% and 99% are also available), using the Wilson formula. Two top-scoring categories are compared using the McNamar test to check if they differ significantly. We also calculate and show Better and Worse coefficients using Timko’s formulas, leaving out reverse and questionable answers. This gives researchers a quick, defensible overview of the data, without hiding ties or using the using unsuitable formulas.

There are several methods for visualizing Kano results, inspired by an informal illustration that Kano provided in the original article to show how fulfilment relates to performance in different categories.

Timko’s coefficients offer a simple way to create a Kano graph. Calculate the “Better” and “Worse” coefficients and then plot features on a Cartesian Kano graph, then compare and rank features visually. Timko placed the two coefficients on a unit square, with Worse along the horizontal axis and Better up the vertical one.

Ting Wang and Ping Ji proposed extending this Kano graph with satisfaction curves. Instead of a unit square, they plot the Better and Worse factors as points on the quadrant axes, and then draw a satisfaction function curve through the two points for visualisation and analysis (line for one-dimensional, a curve that rises quickly and then levels off for must-be, and exponential for attractive).

The curves make the difference between categories easy to see, but their shapes are based on assumptions and not calculations (two points are not enough to determine the full shape of a curve).

DuMouchel’s continuous scoring can also be used to plot items on a Kano graph. Continuous scores then show both the category a feature falls into and how much the respondents agree.

Votito’s survey results page and the free Kano scoring tool show the results both on the 5 × 5 grid of answer pairs, with the number of respondents in each cell, and as a Kano chart of satisfaction curves based on how Wang and Ji visualised the results. We modified the display to follow Kano’s idealised shapes rather than Wang and Ji’s curves, so that the differences between classification types are clearer. If an item can belong to multiple categories, the Kano chart shows all of them, using line style to signal probability (solid, dashed or dotted).

How many people are required for a Kano survey?

Chapman and Callegaro argue that category assignment may be unreliable for samples of fewer than 200 participants. They did find that for larger samples (of more than 1,500 respondents), the category assignments became stable, but even as such they might not be valid for business decision making. (The 200 estimate comes from resampling their research data, evaluating if the smaller sample matched the full sample’s most common category, not whether the assignment was actually correct).

With N < 200, we would expect at least 1/3 of features to be miscategorized more than 20% of the time. With N <= 15 respondents, the overall accuracy rate may be less than 50%.

– Chapman and Callegaro, Kano Analysis: A Critical Survey Science Review

Lee and Newcomb did not provide specific sample size guidelines, but they reported the error margin of their results, so their survey for NASA can provide a rough sense of the scale required for different levels of confidence. With 220 questionnaires returned, they evaluated the precision of a category share at “plus or minus 6 percent with a 90 percent confidence level”, and used that margin as the minimum gap between the two most common categories. The margin of error shrinks only with the square root of the sample size, so halving it to 3 points would need about four times as many respondents.

Kano surveys can be used for systematic quantitative research with large populations of users. The questions and responses are closed (customers can only select from predetermined options) so the responses can be easily analysed at scale. The survey can be used somewhere in the middle of longer research, after collecting the initial ideas and before in-depth studies to decide on product features. It is not reliable enough to be the single deciding factor, but it can confirm previous broader ideas and create narrower targets for follow-up studies.

The question pairing helps to spot some contradictory answers, which can be used for further research or eliminated from the final scoring. It can also point to unclear questions that need to be rephrased. There are no public benchmarks for what’s expected in the Questionable category. In Matzler’s ski survey, no requirement had more than 2.5% questionable answers. Chapman and Callegaro reported between 4 and 7 percent.

Mikulić and Prebežac point out that the survey can be used for evaluating non-existing features, although Chapman and Callegaro warn that answers about unfamiliar features tend to be less reliable.

The BBN team that worked on RS/1 reported that “the survey seems to be well suited to defining a new product, but awkward for developing a new release of an existing mature product”. As an example, they ask how to interpret an existing customer of an existing product rating a new feature as “must-be”.

Kano model analysis of customer requirements was common in Quality Function Deployment, but different companies applied different prioritization methods. A review of two decades of Kano research concludes that “there is thus no agreement on how to use different classes of attributes within QFD”.

Chapman and Callegaro suggest that the nice thing about the Kano model is that it puts features on a “2-dimensional landscape that is excellent for framing discussions with product managers, executives, and the like”, at the same time pointing out systematic issues with validity and reliability of popular survey methods, and that the “theory behind the Kano scores is dubious and has little empirical support”, suggesting that the resulting data from surveys isn’t particularly trustworthy.

As such, the Kano model is best used as a basis for a discussion, not as a way to directly make important product decisions.

Kano Model Limitations

The Kano model can capture expectations and provide a simple overview that is easy to understand, and may imply precision, but in reality it’s not precise. Kano modeling should not be used alone to make key decisions about product design. Instead, it should inform further in-depth research. Several factors limit the validity of the results:

Classification changes over time

The Kano model provides a snapshot of current customer expectations, and the categorization is expected to change over time. What was once exciting becomes normal and expected. In 2001 Kano proposed several life-cycles as a hypothesis. In one, attributes move from indifferent to attractive, then to performance and finally to expected. Kano presents TV remote controls as an example: in surveys in 1983, 1989 and 1998, most people rated it attractive, then nearly half rated it one-dimensional, then more than half rated it must-be. He reported these results only in words, with no sample sizes. The same paper reports a counter-example, with TV voice controls. The opinion shifted from indifferent to one-dimensional and back to indifferent, which Kano says his hypothesis “cannot explain”.

At the time when this article was written, there was no set model for the lifecycle transitions or timelines. Lars Witell and colleagues published a review of 147 Kano studies in 2013, and concluded that “there is basically no knowledge about alternative life cycles, or how many attributes follow these types of life cycles”.

The practical research or academic proof for life-cycles is very thin. Lars Witell and Martin Löfgren published a report with Swedish consumers about 24 packaging attributes in 2003 and in 2009, with different respondents each time, showed that 16 attributes “experienced no change in the overall classification”. Seven of the eight changes began or ended in a “combination”, where the two most common categories did not differ significantly. The authors conclude that their “empirical investigation was not able to verify the existence of the whole life cycle”.

One category rarely fits all customers

A Kano category describes how a group of customers sees a feature, and groups can differ. In a Kano model example from Kano’s own table-clock survey, the size of the clock was rated attractive by people who owned small clocks, but one-dimensional by the rest. When two categories have a similar number of responses, the survey might be “dealing with two market segments”.

Even within one group, the most common category can represent a minority. In Chapman and Callegaro’s data, one feature was classified the same way in five samples, “yet 67–72% of users assigned it to a different category in each sample”. Chapman and Callegaro suggest reporting how answers split across all the categories, not only the majority choice.

A single person might fit into multiple categories

Chapman and Callegaro warn that people might experience several different attitudes about the same function or feature, so assigning just one is always going to be incomplete. They use an example of introducing a new, higher-performance cable for an existing tablet, where the same person might find an argument to support all five attitudes (liking it because it’s faster, expecting it because manufacturers make such changes all the time, tolerating it because they can buy new cables, disliking it because they would prefer not to switch, or being neutral because they can see both pros and cons of the new cable). The rating categories are usually presented as mutually exclusive, but they are not in reality. Asking 1,501 people the same Kano item twice, seconds apart, about a single feature, Chapman and Callegaro found that the same people would often change choices between different attitudes towards the same question. For example, only 39% of those who first answered “expect” gave the same answer again; 40% changed to “like”.

The model does not consider the magnitude or limits of performance

The model classifies features in one of the categories with no degrees of achievement or of satisfaction, so Kano prioritization cannot differentiate between minor annoyances and huge problems, or equally between slight exciters and something that would be very desirable.

Comparing “attractive” classifications with expected behavioural impact, Benedikt M. Brand and Alexandra Rese surveyed 607 German consumers, and found that many of the presumed “attractive” attributes do not increase the visiting frequency for stores, so it would be wrong to make product decisions just based on classification.

Kano’s model also has no way to express when performance and exciter aspects surpass the ability of the market to notice or consume further improvements. At some point, more is no longer better, as the customers will not notice any significant change to their work. The QUPER model breakpoints can be used to complement Kano model analysis and establish when performance enhancements start to bring diminishing returns.

The survey is unreliable

The Kano survey suffers from several structural issues that make it unreliable. In a study that compared Kano question formats against direct customer classification, Witell and Löfgren found that the results disagreed on almost every attribute.

Using more than 10,000 respondents, Chapman and Callegaro evaluated the reliability and validity of seven survey versions that used the same items, but differed in details such as item order. They started with three features that would be expected to fit clearly into one of the main Kano categories, and found that the results only partially aligned with Kano theory. Giving respondents more freedom in a version that allowed people to select multiple options, Chapman and Callegaro concluded that the pattern of item associations was “largely inconsistent with Kano theory and assumptions”, apart from the “Dislike” answers. Even more curiously, in a retest, they discovered that 61% of the people who chose “expect” chose something else the second time.

As potential reasons for such unreliability, Chapman and Callegaro suggest that people are being asked about hypothetical attitudes, which makes answering some questions difficult, in particular asking how people would feel about having a function that’s not yet introduced into a product.

Chapman and Callegaro suggest that the result scale “conflates multiple dimensions”, but is usually presented as a single, nominal scale. Using exploratory factor analysis on the correlations between answers in versions where users could select multiple options, Chapman and Callegaro argue that the scale conflates at least 2, and possibly 4 or 5, dimensions. Presenting them on a single-dimensional scale is misleading.

Lars Witell and Martin Löfgren compared the standard five-level questionnaire with a three-level version, and two other ways of ranking quality attributes for an online ticket sales service, found wide disagreement between the options: none of the alternatives agreed with the standard questionnaire on more than two of nine attributes. Li-Fei Chen compared a Kano-style questionnaire with methods that infer the categories from satisfaction ratings in a study of 210 restaurant customers, with the best match only agreeing on 18 of 28 attributes, and the worst agreeing only on one.

The model might expect durable properties

Fast-moving markets with frequent improvements may not be a good fit for Kano classification. Chapman and Callegaro argue that the concepts “are based on durable consumer goods”, and “may be inapplicable for technology products” where customers are used to constant improvement.

The BBN team also notes that the survey was not very useful for distinguishing between small features and large ones.

Chapman and Callegaro warn that the Kano method gives a “compelling” answer to questions about features, but that “it will tell a story—quite possibly an incorrect story”.